Your prompt, with the rules already attached.

Type what you would ask any AI. Before it is answered, 32 operating rules for an agent are placed above it: verify before claiming, answer every part, never repeat a corrected mistake, close out as DONE / YOUR MOVE. Use the answer here, or take the enriched prompt to ChatGPT, Claude, Gemini or Grok.

1. Ask here The hosted model gets the full private rulebook (133 sections) and was trained on 1,096 corrections that stuck. checking…

2. Copy the enriched prompt Same box, second button. Paste the result into any AI you already use. Nothing to install, nothing stored.

3. Call it as an API OpenAI-compatible. Point any tool at this address and every request is enriched before the model sees it. Details below.

See the 32 rules that get added

loading…

As an API

Drop-in for anything that speaks the OpenAI chat format. Keys are minted per integration; write to [email protected] for one.

curl https://model.howardchan.me/v1/chat/completions \
  -H "x-api-key: YOUR_KEY" -H "content-type: application/json" \
  -d '{"messages":[{"role":"user","content":"Close out this task."}]}'

# only the enriched prompt, no model call, no key:
curl https://model.howardchan.me/api/enrich -d '{"prompt":"Close out this task."}'

In Codex CLI: a provider with base URL https://model.howardchan.me/v1 and the key in its env var. In Claude Code: the same address through claude-code-router.

Runs on one founder's own machine, behind Cloudflare. Rules written over five months of running five companies through AI agents; the public set is the part anyone can use. Howard Chan, howardchan.me. Related: Does your AI listen?